Contrastive Analysis of Deep Learning-Based Urban Heat Island Superresolution Modeling with Land Use Embeddings

Keywords: Urban Heat Island, deep learning, superresolution
Abstract Type: Paper Abstract

Abstract

Urban heat islands are regions within the urban areas where temperatures are significantly higher than in the surrounding environments due to disparities in surface albedo and retaining higher energy from solar radiation. In this research we analyze defining evolution of heat islands in Chicago, Illinois from a deep learning-based superresolution model combining DoE’s E3SM with a dense urban weather station array. The resulting spatiotemporal series is analyzed with graph-based regionalization to determine data-driven heat islands. By utilizing high-resolution temperature data alongside land use information, the proposed model produces temperature variations at high spatial resolution, while contrastive learning enables learning urban heat islands against a changing baseline of temperature. Results show strong collocation of heat islands with areas away from greenspaces, emphasizing the solutions to improve urban thermal environment.